Researchers have developed a new reasoning framework, Reasoning via Planning (RAP), to enhance the reasoning capabilities of Large Language Models (LLMs). RAP uses a library of advanced reasoning algorithms, treating multi-step reasoning as planning. It optimizes the balance between exploration and exploitation using the concepts of ‘World Model’ and ‘Reward’. The framework has shown superior performance over several contemporary reasoning approaches, even outperforming GPT-4 in certain settings. This innovative combination of planning and reasoning could revolutionize the approach to LLM reasoning, potentially enabling AI systems to achieve human-level strategic thinking and planning.
Read more at MarkTechPost…
